Models / Llama 3.3 70B Instruct

Llama 3.3 70B Instruct

enthusiast
HuggingFace → Meta → Find for my hardware

AI-generated content marks

The provider reports that this model does not add embedded watermarks or provenance metadata to generated output.

Provider transparency docs →

coding reasoning writing chat
Parameters
70.6B
Context
128k
License
llama 3
Developer
Meta
Origin
🇺🇸 USA
Released
Dec 2024

Scores

Coding
72
Reasoning
77
General
80

Score per dollar

800 pts per $/M input

general_score (80) divided by cheapest input price ($0.10/M). Higher is better value. See live pricing.

Related models

Llama 3.2 1B Instruct
1.2B edge 1GB min RAM $0.03/M in Sep 25, 2024
Llama 3.2 3B Instruct
3.2B edge 2GB min RAM $0.05/M in Sep 25, 2024
Llama 3.1 405B
405.0B premier no local build Jul 23, 2024
Llama 3.1 8B Instruct
8.0B consumer 5GB min RAM $0.02/M in Jul 23, 2024

Guides covering Llama 3.3 70B Instruct

How AT&T cut AI coding costs 56% with open-weight models and LiteLLM Aug 28, 2026

Save your hardware and every model page answers the real question: will it run on your machine, and how fast?

Join free - save your rig →

Run it locally

Per-quant memory needs and a static "can you run it?" reference - no rig entry required

Q2_K
21.5GB 21.5GB min 25.0GB rec
Smallest footprint, noticeable quality loss
Q3_K_M
30.0GB 30.0GB min 34.0GB rec
Compact, moderate quality loss
Q4_K_M
42.0GB 42.0GB min 48.0GB rec
Balanced - the usual local sweet spot
Q8_0
74.0GB 74.0GB min 80.0GB rec
Near-lossless

The reference hardware

Schematic of the NVIDIA Jetson Orin NX 16GB reference rig - 16GB unified memory, 102 GB/s aggregate bandwidth
Schematic of the Single GTX 1080 Ti (11GB) reference rig - 11GB VRAM, 484 GB/s aggregate bandwidth
Schematic of the Single RTX 4090 (24GB) reference rig - 24GB VRAM, 1008 GB/s aggregate bandwidth
Schematic of the 4x H100 80GB (320GB) reference rig - 320GB VRAM, 13400 GB/s aggregate bandwidth
Schematic of the NVIDIA DGX Station 748GB reference rig - 748GB unified memory, 8000 GB/s aggregate bandwidth
Schematic of the 8x RTX 3090 rack (192GB) reference rig - 192GB VRAM, 7489 GB/s aggregate bandwidth
Schematic of the 4x RTX 5090 (128GB) reference rig - 128GB VRAM, 7168 GB/s aggregate bandwidth
Schematic of the AMD Instinct MI300X (192GB) reference rig - 192GB VRAM, 5324 GB/s aggregate bandwidth
Schematic of the 4x RTX 4090 (96GB) reference rig - 96GB VRAM, 4032 GB/s aggregate bandwidth
Schematic of the 2x RTX 5090 (64GB) reference rig - 64GB VRAM, 3584 GB/s aggregate bandwidth
Schematic of the 2x RTX 3090 (48GB) reference rig - 48GB VRAM, 1872 GB/s aggregate bandwidth
Schematic of the Single RTX 5090 (32GB) reference rig - 32GB VRAM, 1792 GB/s aggregate bandwidth
Schematic of the RTX PRO 6000 Blackwell (96GB) reference rig - 96GB VRAM, 1792 GB/s aggregate bandwidth
Schematic of the Mac Studio M4 Ultra 192GB reference rig - 192GB unified memory, 1092 GB/s aggregate bandwidth
Schematic of the Mac Studio M4 Ultra 512GB reference rig - 512GB unified memory, 1092 GB/s aggregate bandwidth
Schematic of the MacBook Pro M5 Max 128GB reference rig - 128GB unified memory, 614 GB/s aggregate bandwidth
Schematic of the Dual EPYC 9004 + 768GB DDR5-4800 reference rig - 768GB unified memory, 460 GB/s aggregate bandwidth
Schematic of the DGX Spark 128GB unified reference rig - 128GB unified memory, 273 GB/s aggregate bandwidth
Schematic of the Ryzen AI Max+ 395 128GB reference rig - 128GB unified memory, 256 GB/s aggregate bandwidth
Schematic of the Jetson AGX Orin 64GB reference rig - 64GB unified memory, 204 GB/s aggregate bandwidth
Schematic of the Epyc + 512GB DDR4-3200 + 2x RTX 3090 reference rig - 560GB unified memory, 204 GB/s aggregate bandwidth
Schematic of the Epyc + 512GB DDR4-2400 + 2x RTX 3090 reference rig - 560GB unified memory, 153 GB/s aggregate bandwidth

22 reference configs, drawn in-house. Scroll for more.

Can you run it? - reference rigs

Rig Q2_K Q3_K_M Q4_K_M Q8_0
NVIDIA Jetson Orin NX 16GB no -> cloud no -> cloud no -> cloud no -> cloud
Single GTX 1080 Ti (11GB) no -> cloud no -> cloud no -> cloud no -> cloud
Single RTX 4090 (24GB) tight offload no -> cloud no -> cloud
4x H100 80GB (320GB) fast 342.7t/s fast 245.6t/s fast 175.5t/s fast 99.6t/s
NVIDIA DGX Station 748GB fast 204.6t/s fast 146.6t/s fast 104.8t/s fast 59.5t/s
8x RTX 3090 rack (192GB) fast 191.5t/s fast 137.3t/s fast 98.1t/s fast 55.7t/s
4x RTX 5090 (128GB) fast 183.3t/s fast 131.4t/s fast 93.9t/s fast 53.3t/s
AMD Instinct MI300X (192GB) fast 136.2t/s fast 97.6t/s fast 69.7t/s fast 39.6t/s
4x RTX 4090 (96GB) fast 103.1t/s fast 73.9t/s fast 52.8t/s fast 30.0t/s
2x RTX 5090 (64GB) fast 91.7t/s fast 65.7t/s fast 46.9t/s offload
2x RTX 3090 (48GB) fast 47.9t/s fast 34.3t/s fast 24.5t/s offload
Single RTX 5090 (32GB) fast 45.8t/s tight offload no -> cloud
RTX PRO 6000 Blackwell (96GB) fast 45.8t/s fast 32.9t/s fast 23.5t/s ok 13.3t/s
Mac Studio M4 Ultra 192GB fast 30.5t/s fast 21.8t/s ok 15.6t/s ok 8.9t/s
Mac Studio M4 Ultra 512GB fast 30.5t/s fast 21.8t/s ok 15.6t/s ok 8.9t/s
MacBook Pro M5 Max 128GB ok 17.1t/s ok 12.3t/s ok 8.8t/s slow 5.0t/s
Dual EPYC 9004 + 768GB DDR5-4800 ok 11.8t/s ok 8.5t/s slow 6.0t/s slow 3.4t/s
DGX Spark 128GB unified slow 7.0t/s slow 5.0t/s slow 3.6t/s slow 2.0t/s
Ryzen AI Max+ 395 128GB slow 6.6t/s slow 4.7t/s slow 3.4t/s slow 1.9t/s
Jetson AGX Orin 64GB slow 5.2t/s slow 3.8t/s slow 2.7t/s no -> cloud
Epyc + 512GB DDR4-3200 + 2x RTX 3090 slow 5.2t/s slow 3.8t/s slow 2.7t/s slow 1.5t/s
Epyc + 512GB DDR4-2400 + 2x RTX 3090 slow 3.9t/s slow 2.8t/s slow 2.0t/s slow 1.1t/s

Fit tiers use the same will-it-run logic as the rig finder. For comfortable fits, the badge reflects decode speed: fast >=20 t/s, ok 8-20 t/s, slow <8 t/s. t/s is a bandwidth estimate, not a measured benchmark.

Download options

Q2_K official -20% vs fp16
21.5GB dl 21.5GB min 25.0GB rec
REC RAM vs largest quant
Q3_K_M official -10% vs fp16
30.0GB dl 30.0GB min 34.0GB rec
REC RAM vs largest quant
Q4_K_M official -4% vs fp16
42.0GB dl 42.0GB min 48.0GB rec
REC RAM vs largest quant
Q8_0 official -1% vs fp16
74.0GB dl 74.0GB min 80.0GB rec
REC RAM vs largest quant

Or run it in the cloud

Live per-provider pricing, throughput and uptime - refreshed about 9 hours ago via OpenRouter. Click a column to sort.

some pricing may be stale - last verified 2026-09-29

Provider Type Input $/M Output $/M Cache $/M Tok/s Latency Uptime Value
Google
API 0.72 0.72 - - - -
Together AI stale
API 0.88 0.88 - - - -
Fireworks AI stale
API 0.90 0.90 - - - -
Together
API 1.04 1.04 - - - 100.00% best uptime
CoreWeave
API 0.71 0.71 0.710 - - 99.92%
Groq
API 0.59 0.79 0.295 - - 99.88%
Novita
API 0.14 0.40 - - - 99.85%
SambaNova
API 0.45 0.90 - - - 99.59%
AkashML
API 0.20 0.52 0.100 - - 99.49%
DeepInfra
API 0.10 0.32 - - - 99.36% cheapest
Parasail
API 0.22 0.50 0.110 - - 98.32%
Cloudflare
API 0.29 2.25 - - - 95.99%

Default order: throughput among 95%+ uptime providers, then latency; subscriptions last. Sort by any column. Subscription rows show $/mo in the Value column - per-token columns are "-". Affiliate links are marked sponsored / nofollow. Confirm current pricing on the provider's site before committing.

Detailed API pricing page + JSON endpoint →

See who runs Meta in production →

PRICE HISTORY

Inference cost over time

Data accumulates from the first daily sync - longer ranges populate over time. Prices come from OpenRouter snapshots, not a historical API.

Loading price history...